The Content Optimizer runs a multi-round Bayesian optimization loop where a seeder proposes strategic ranking hypotheses, an ideator crafts localized snippet or page edits, and an LLM ranker scores the variants against a competitive set to update posterior beliefs.